arX

iv:0

901.

2046

v1 [

astr

o-ph

.GA

] 1

4 Ja

n 20

09to appear in the Astrophysical JournalPreprint typeset using LATEX style emulateapj v. 12/14/05

REST-FRAME UV VERSUS OPTICAL MORPHOLOGIES OF GALAXIES USING SERSIC PROFILE FITTING:THE IMPORTANCE OF MORPHOLOGICAL K-CORRECTION

Abhishek Rawat,1 Yogesh Wadadekar,2 Duilia De Mello,3,4,5

to appear in the Astrophysical Journal

ABSTRACT

We show a comparison of the rest-frame UV morphologies of a sample of 162 intermediate redshift(zmedian = 1.02) galaxies with their rest-frame optical morphologies. We select our sample fromthe deepest near-UV image obtained with the Hubble Space Telescope (HST) using the WFPC2(F300W) as part of the parallel observations of the Hubble Ultra Deep Field campaign overlapping withthe HST/ACS GOODS dataset. We perform single component Sersic fits in both WFPC2/F300W(rest-frame UV) and ACS/F850LP (rest-frame optical) bands and deduce that the Sersic index nis estimated to be smaller in the rest-frame UV compared to the rest-frame optical, leading to anoverestimation of the number of merger candidates by ∼40%-100% compared to the rest-frame opticaldepending upon the cutoff in n employed for identifying merger candidates. This effect seems to bedominated by galaxies with low values of n(F300W ) ≤ 0.5 that have a value of n(F850LP ) ∼ 1.0. Weargue that these objects are probably clumpy starforming galaxies or minor mergers, both of whichare essentially contaminants, if one is interested in identifying major mergers.In addition we also find evidence that the axis ratio b/a is lower, i.e. ellipticity (1− b/a) is higher in

rest-frame UV compared to the rest-frame optical. Moreover, we find that in the rest-frame UV, thenumber of high ellipticity (e ≥ 0.8) objects are higher by a factor of ∼ 2.8 compared to the rest-frameoptical. This indicates that the reported dominance of elongated morphologies among high-z LBGsmight just be a bias related to the use of rest-frame UV datasets in high-z studies.Subject headings: galaxies: evolution - galaxies: formation - galaxies: statistics

1. INTRODUCTION

Most of our knowledge about the properties of highredshift galaxies has been acquired by studying LymanBreak Galaxies (LBGs) at redshift ∼2.0-5.0 (e.g. Gi-avalisco 2002; Steidel et al. 2003; Ravindranath et. al.2006; Wadadekar, Casertano & de Mello 2006). Sincemost of these studies are in the optical, they probe therest-frame UV light of LBGs at these redshifts. Deriv-ing morphological information about a galaxy from itsrest-frame UV light can be misleading as the UV lightis mainly dominated by patchy star forming regions thatare not representative of the underlying galaxy mass dis-tribution (e.g. Teplitz et al. 2006). Even so, people haveused a variety of techniques to derive the morphologies ofsuch high-z objects, including the non-parametric CAS(Conselice et al. 2003) and Gini/M20 coefficients (Lotzet al. 2004) and the parametric light profile fitting suchas Galfit (Peng et al. 2002). While each of these meth-ods have their pros and cons, all of them have been usedin the past (and continue to be used) by the communityfor studying quantitative morphologies of high redshiftgalaxies. In light of this fact, it becomes imperative tobe aware of the biases that are expected, as one probesprogressively bluer rest-frame wavelengths at higher red-shifts using each of these techniques.Papovich et al. (2005) and Conselice et al. (2005)

1 Inter University Centre for Astronomy and Astrophysics, PostBag 4, Ganeshkhind, Pune 411007, India: rawat@iucaa.ernet.in

2 National Centre for Radio Astrophysics, Post Bag 3,Ganeshkhind, Pune 411007, India.

3 Observational Cosmology Laboratory, Code 665, GoddardSpace Flight Center, Greenbelt, MD 20771, USA

4 Catholic University of America, Washington, DC 20064, USA5 Johns Hopkins University, Baltimore, MD 21218, USA

have carefully examined the extent of this morphologi-cal K-correction using WFPC2 and NICMOS observa-tions of the Hubble Deep Field North (HDF-N). Morerecently, Conselice et al. (2008) have examined the effectsof morphological K-correction on the estimation of non-parametric CAS (Conselice et al. 2003) and Gini/M20coefficients (Lotz et al. 2004), using ACS and NICMOSimaging of the Hubble Ultra Deep Field. They find thatCAS and Gini/M20 parameters give similar results, ex-cept that the latter tend to find larger number of mergersthan the former. Despite the fact that Galfit has beenused extensively for quantifying galaxy morphologies athigh redshifts (eg. Ravindranath et al. 2006), there hasbeen no systematic study evaluating the effects of mor-phological K-correction on the structural parameters de-rived using Galfit. It is this problem that we are trying toaddress in the current paper. In this work, we quantifythe biases that are encountered in estimating the mor-phological parameters of galaxies in the rest-frame UVas compared to rest-frame optical light using Galfit, astructural decomposition code that fits 2-D parameter-ized, axisymmetric, functions directly to galaxy images.

2. THE DATA AND SAMPLE SELECTION

We have utilized the HST/WFPC2 F300W band ob-servations (orients 310 and 314) obtained in parallel withthe HST/ACS HUDF that fall within the GOODS-S areaand constitute the deepest image of the near UV sky everobtained with the HST (total exposure ∼323.1 ksec).The dataset consists of a total of 409 WFPC2 F300Wparallel images, with exposure times ranging from 700sec - 900 sec each. A cosmic-ray-rejected, drizzled imagewith a pixel scale of 0.06 arcsec/pixel was constructedby combining the 409 individual exposures (de Mello et

2

Fig. 1.— The photometric redshift distribution of our 162 U-band selected galaxies with optical detection (zmag ≤ 25.5). Themedian photometric redshift is 1.02.

al. 2006; Wadadekar, Casertano & de Mello 2006), usinga drizzle-based technique developed for data processingby the WFPC2 Archival Parallels Project (Wadadekaret al. 2006). This drizzled image, which was accuratelyregistered with respect to the GOODS images, was thenused for performing the morphological analysis presentedin this paper.For the optical imaging, we have used the publicly

available version v1.0 of the reduced, calibrated imagesof the Chandra Deep Field South (CDFS) acquired withHST/ACS as part of the Great Observatories OriginsDeep Survey, GOODS (Giavalisco et al. 2004). The over-lap between the HST/WFPC2 F300W image footprintand the HST/ACS GOODS image footprint is shown inFig. 1 of de Mello et al. (2006).A SExtractor (Bertin & Arnouts 1996) based source

catalog was prepared for the WFPC2/F300W image byde Mello et al. (2006) containing 415 sources. We po-sitionally matched this F300W-band based catalog withthe publicly available HST/ACS version r1.1 multi-bandsource catalog to find optical counterparts for the 415 UVbright sources. We only included optical counterparts ofUV sources brighter than zmag = 25.5 and having a pho-tometric redshift (photo-z) as determined by de Mello etal. (2006). This yielded a sample of 162 sources havingphotometric redshift distribution as shown in Fig. 1

3. THE METHODOLOGY

Our aim is to quantify any biases in the derived galfitparameters, on account of using rest-frame UV light ofprogram objects, as opposed to their rest-frame opticallight. In order to do this, we performed single componentSersic fits to our sample of 162 objects in both UV as wellas optical. The goal is to look for systematic differencesin the derived galaxy parameters as a function of theirrest-frame wavelength.

3.1. The F300W band fits

We used the software Galfit to carry out two-dimensional modeling for our sample of galaxies. Weperformed single component Sersic fits, where the inten-sity profile of the galaxy is modeled with the Sersic law

(Sersic 1968),

Σ(r) = Σeexp[

−2.303bn

[( r

re

)1/n

− 1]]

(1)

with re being the half light radius, Σe the surfacebrightness at re and n being the Sersic index and bnis related to n such that P (2n, 2.303bn) = 0.5, whereP (a, b) is the incomplete gamma function.The model galaxy is constructed as a combination of a

Sersic component and the sky background value. Otherfree parameters in the model are the ellipticity of theSersic component and its position angle. The SExtrac-tor based photometric catalog of our sources provided uswith valuable starting values for most of the above men-tioned parameters. This analytic model of the galaxylight distribution is then convolved with the Point SpreadFunction (PSF) of the observation and compared withthe observed galaxy image. The values of the free pa-rameters are determined iteratively by minimizing thereduced χ2. In Fig. 2 we show an example fit for one ofthe targets in the F300W band.In the above scheme, an accurate estimation of the

PSF is crucial to the robust determination of the struc-tural parameters of a galaxy. However, there are only 4identifiable stars in the F300W image which can be usedto characterize the PSF. One of these was saturated andanother was too close to the edge of the frame. We usedthe other two stars (say X and Y) in turn, to establishthat the values of the derived parameters are not severelydependent on the choice of the PSF star.Here we use the methodology of Rawat et al. (2007) to

show that the derived parameters are invariant under thechange of the PSF star. For our sample of 162 objectsin the F300W band, we used Galfit in batch mode toderive the structural parameters twice for each object.In the first run, we used one of the stars (X) as thePSF. In the second run, we used the second star (Y)as the PSF, everything else remaining the same as in thefirst run. In Fig. 3 we show the difference between thestructural parameters (n and b/a) obtained in the twocases as a function of the F300W band magnitude of theobjects. As is evident, our estimation of the structuralparameters is quite robust irrespective of which star isused as the PSF. The derived structural parameters inthe two runs agree quite well for most of the objects,with a symmetrical scatter around the zero line, pointingtowards random errors as the cause for any dispersionin the structural parameters obtained in the two runsrather than any systematic effect. For fainter objects,the random photon noise is higher and may overwhelmany systematic effect caused by the use of two differentPSF stars. Also, the Sersic index n, which is well knownto be difficult to constrain, exhibits a greater amount ofscatter than the axis ratio b/a. The dashed lines in thetwo panels encompass ∼ 80% of all the objects.

3.2. The F850LP band fits

Single component Sersic fits were performed on theHST/ACS F850LP image of the 162 objects. The fitswere performed in a manner consistent with the F300W-band fitting procedure. Since in the z-band we have alarge number of identifiable stars which can be used forcharacterizing the PSF, we have used the star closest to

3

-3 -2 -1 0 1 2 3

-3

-2

-1

0

1

2

3

J033239.97-274213.6 0.62

0.24 25.55

-3 -2 -1 0 1 2 3

-3

-2

-1

0

1

2

3

-3 -2 -1 0 1 2 3

-3

-2

-1

0

1

2

3

Fig. 2.— An example of single component Sersic fit in the F300W band. From left to right are shown the F300W band image, the GalfitSersic best fit model and the residual after subtracting the model from the image. The IAU name of the object is indicated at top left, thephotometric redshift of the object is given at top right, the fitted Sersic index n is given at bottom left and the magnitude of the object isgiven at bottom right of the object frame. Each image is 6 arcsec × 6 arcsec.

Fig. 3.— The difference between the structural parameters nand b/a obtained in two separate Galfit runs, using two differentstars as the PSF in the F300W band as a function of the F300Wmagnitude of the objects. As expected, the difference flares out atfainter magnitudes. The dashed lines in the two plots encompass∼ 80% of all the objects.

a given galaxy as the model PSF. This choice is not crit-ical; as demonstrated by Rawat et al. (2007) the derivedGalfit parameters are stable to within 10% irrespectiveof which star is used as the PSF. One of the fits that weobtained in the F850LP band is shown in Fig. 4 for anobject at photo-z=0.98.

4. COMPARISON BETWEEN F300W AND F850LP FITS

Fig. 5 shows some examples of the morphological dif-ferences in the appearance of a galaxy observed in theUV in comparison to its appearance in the optical. Ineach case, the left panel is a F300W HST/WFPC2 im-age while the right panel is an F850LP HST/ACS image.The first object is a red, edge on disk galaxy. The sec-ond and third objects are classified as mergers (n ≤ 0.8)in the UV (see below) but not in the optical. The lastobject is classified as merger in the UV, but shows clearspiral structure in the optical.Before we compare the derived quantitative param-

eters for our 162 objects in the F300W and F850LPband, it is important to check for possible biases dueto the varying Signal-to-Noise Ratio (SNR) in the twofilter datasets. The F300W band observation suffersfrom the poor UV sensitivity of the WFPC2 detector(∼1.9% for WFPC2/F300W filter)1, while the F850LP

1 Efficiency near filter pivot wavelength, includingHST+instrument+filter as quoted from Table 1.2:Compari-son of WFPC2 and ACS Filters of WFPC2 Instrument Handbookv9.0

band observations gain from the exquisite red sensitiv-ity of the ACS/WFC detector (∼25% for ACS/WFCF850LP filter)1. This vast difference in the UV vs opticalsenstitvities is compensated to a large extent by the ex-ceptionally deep exposure in the WFPC2/F300W (∼ 323ksec), compared to ∼10 ksec in the ACS/F850LP. In or-der to quantify the expected SNR of a typical object inthe UV vs optical, we made use of the exposure timecalculators for the two detectors2,3. The median surfacebrightness for our 162 sample objects in the F300W bandis 24.03 mag/arcsec2 compared to 22.95 mag/arcsec2 inthe F850LP band. We used the known exposure timesin each of the two filters to estimate the expected SNRfor the above median surface brightness objects using theHST/ACS and WFPC2 exposure time calculators, yield-ing an average SNR of ∼21 per pixel in the F300W bandand ∼73 per pixel in the F850LP band.Ravindranath et al. (2006) have performed extensive

Monte Carlo simulations to estimate the reliability ofthe morphological parameters derived from 2D galaxyfitting algorithm Galfit, as a function of the SNR of thegalaxy images. Their simulations suggest that the mor-phological parameters can be well recovered for objectswith SNR≥10-15. Since we have used the methodologysimilar to that of Ravindranath et al. (2006) for deriv-ing morphological parameters and the same code (Galfit)for our sample galaxies, we believe that even though theSNR in the F300W band images for our sample objectsis significantly lower compared to the F850LP band, it isstill sufficient (SNR≥10-15) to allow an accurate estima-tion of their structural parameters. Therefore, we do notexpect any bias in the derived structural parameters inthe F300W band compared to F850LP band due to thedifference in their SNR.Fig. 6 (left) shows the distribution of the derived Sersic

index n for the 162 sources in the F300W band (450 hash-ing) as well as the F850LP band (1350 hashing). As isapparent, the two distributions are quite different, withthe distribution in F300W band peaking at much smallervalues of n, compared to the F850LP band. A two sam-ple KS test confirms that there is > 99% probability

2 WFPC2 ETC for Extended Sources - v4.0http://www.stsci.edu/hst/wfpc2/software/wfpc2-etc-extended-source-v40.html

3 ACS Imaging ETChttp://etc.stsci.edu/webetc/acsImagingETC.jsp

4

-3 -2 -1 0 1 2 3

-3

-2

-1

0

1

2

3

J033233.40-274138.9 0.98

1.91 21.69

-3 -2 -1 0 1 2 3

-3

-2

-1

0

1

2

3

-3 -2 -1 0 1 2 3

-3

-2

-1

0

1

2

3

Fig. 4.— An example of single component Sersic fit in the F850LP band. From left to right are shown the F850LP band image, theGalfit Sersic best fit model and the residual after subtracting the model from the image. Labels are same as in Fig. 2. Each image is 6arcsec × 6 arcsec

of the two samples being drawn from a different parentpopulation. Fig. 6 (right) shows a scatter plot of Sersicindex n(F300W ) vs n(F850LP ). Note the clumping atthe bottom left, showing objects with n(F300W ) ≤ 0.5and n(F850LP ) ∼ 1.0, corresponding to the peaks in theSersic index distribution at n ≤ 0.5 in the F300W bandand n ∼ 1.0 in the F850LP band. The average errorbarshave been derived using the differences in the computedSersic index n using two different PSFs in the F300Wband as shown in Fig. 3.We notice that in the F300W band, the Sersic index

n is lower in a large number of cases compared the theF850LP band. This effect is mainly caused by galaxieswith low values of n(F300W ) ≤ 0.5 that have a cor-responding n(F850LP ) ∼ 1.0, while this effect does notseem to be important in galaxies with high values of n, ascan be clearly seen in Fig. 6(right). This has serious con-sequences, as n is widely used in the literature for iden-tifying the morphological class of a galaxy. Especially,n flatter than exponential is usually used for isolatinggalaxies with multiple cores or disturbed morphologies,possibly indicative of mergers (eg. Ravindranath et al.2006). However, it is imperative to note here that itis impossible to isolate a clean sample of merger candi-dates using this criterion, as the objects that exhibit nshallower than exponential profile include galaxies withpatchy star forming HII regions and low surface bright-ness galaxies etc., in addition to true merging galaxies(see eg. the discussion in Ravindranath et al. 2006).Keeping the above mentioned caveat in mind, we findfrom Fig. 6 (left) that nF300W ≤ 0.5 for 57 objects,whereas nF850LP ≤ 0.5 for only 28 cases. This meansthat the use of F300W band, which probes the rest-frameUV light of the objects at these redshifts, increases thenumber of merger candidates by a factor ∼ 2 comparedto the F850LP band (rest-frame optical). This factor isa less severe (but no less significant) ∼ 1.4, if we employa n ≤ 0.8 as the cutoff for identifying merger candidates.We checked whether the objects with n ≤ 0.5 in ei-

ther of the two filters are systematically fainter than thewhole sample, i.e. whether poorer SNR might be re-sponsible for these objects having anomalously low n.We found that in the F300W band, the median surfacebrightness for the subset of 57 objects with nF300W ≤ 0.5is 24.08 mag/arcsec2 (compared to 24.03 mag/arcsec2

for the whole sample). This yields a SNR of ∼20 per

pixel (compared to ∼21 for the whole sample). Simi-larly, in the F850LP band, the median surface bright-ness for the subset of 28 objects with nF850LP ≤ 0.5 is23.05 mag/arcsec2 (compared to 22.95 mag/arcsec2 forthe whole sample). This yields an SNR of ∼67 per pixel(compared to ∼73 per pixel for the whole sample). Hencewe see that the SNR for the subset of objects with ex-tremely low values of n in both the filters is very similarto that for the entire sample indicating that the param-eters derived for these objects are likely to be just asrobust as for the entire sample.The fact that the differences in the derived value of

n in the two filters are dominated by galaxies withlow values of n(F300W ) ≤ 0.5 that have a value ofn(F850LP ) ∼ 1.0 leads us to believe that we might prob-ing clumpy star forming HII regions in these objects inthe F300W band. The tacit assumption that we aremaking here is that clumpy star formation in galaxieswill lead to a lower n value in the rest-frame UV butnot in the rest-frame optical (because star forming HIIregions are blue). This would mean that objects withn(F300W ) ≤ 0.5 and n(F850LP ) ∼ 1.0 are probablyclumpy starforming galaxies which are intrinsically disky(n ∼ 1.0). While it is true that blue neighbours will tendto mimic HII regions in the rest-frame UV by producinga shallow Sersic profile, however the very fact that theycontribute little or no flux in the optical (parametrized bya steeper n(F850LP ) ∼ 1.0) means that they cannot bemassive systems (as z band flux is a better tracer of stel-lar mass than UV flux) and will constituteminor mergersat best (see Rawat et al. (2008) for further discussion onhow flux in redder wavebands is better suited for identi-fying massive major merger candidates). So objects withn(F300W ) ≤ 0.5 and n(F850LP ) ∼ 1.0 are likely to beeither clumpy star forming galaxies or minor mergers,both of which are essentially contaminants, if one is in-terested in identifying major mergers. In Section 5, weinclude comments on the morphology of individual ob-jects with n(F300W ) ≤ 0.5 and 0.7 ≤ n(F850LP ) ≤ 1.3(the working definition of n(F850LP ) ∼ 1.0) in detail toshow that our assertion above is indeed correct.Similarly, Fig. 7 (left) shows the distribution of the

axis ratio (b/a) in the F300W band (450 hashing) aswell as in the F850LP band (1350 hashing). From thefigure, we find that (b/a)F300W ≤ 0.2 for 34 objects,while (b/a)F850LP ≤ 0.2 for only 12 objects. This means

5

-3 -2 -1 0 1 2 3

-3

-2

-1

0

1

2

3

J033234.61-274224.1 0.42

1.76 25.52

-3 -2 -1 0 1 2 3

-3

-2

-1

0

1

2

3

J033234.61-274224.1 0.42

1.77 20.94

-3 -2 -1 0 1 2 3

-3

-2

-1

0

1

2

3

J033235.88-274145.1 0.59

0.74 25.06

-3 -2 -1 0 1 2 3

-3

-2

-1

0

1

2

3

J033235.88-274145.1 0.59

1.10 23.10

-3 -2 -1 0 1 2 3

-3

-2

-1

0

1

2

3

J033239.04-274132.4 0.69

0.16 23.64

-3 -2 -1 0 1 2 3

-3

-2

-1

0

1

2

3

J033239.04-274132.4 0.69

1.03 22.02

-3 -2 -1 0 1 2 3

-3

-2

-1

0

1

2

3

J033236.24-274121.9 0.66

0.19 26.09

-3 -2 -1 0 1 2 3

-3

-2

-1

0

1

2

3

J033236.24-274121.9 0.66

1.70 22.14

Fig. 5.— Some examples of the difference in the appearance of a galaxy observed in the UV (left) in comparison with its appearancein the optical (right). The left cutout is a F300W filter HST/WFPC2 image from the Hubble-UDF parallels, while the right cutout is theF850LP HST/ACS image of the same object from the GOODS survey. Labels are same as in Fig. 2. Each image is 6 arcsec × 6 arcsec.

Fig. 6.— (left) A comparison between the Sersic index n derived using Galfit, single component Sersic fits in the WFPC2/F300W-band(450 hashing) and the ACS/F850LP band (1350 hashing) for our 162 galaxies. A two sample KS test shows that there is > 99% probabilityof the two samples being drawn from a different parent population. (right) A scatter plot of Sersic index n(F300W ) vs n(F850LP ).Note the clumping at the bottom left (enclosed by the black rectangle), showing objects with n(F300W ) ≤ 0.5 and n(F850LP ) ∼ 1.0,corresponding to the peaks at n ≤ 0.5 in the F300W band and n ∼ 1.0 in the F850LP seen in the Sersic index distribution (left). Themorphology of these objects is discussed in Section 5. The average errorbars are derived using the differences in the computed Sersic indexn using two different PSFs in the F300W band as shown in Fig. 3.

6

Fig. 7.— (left) A comparison between the axis-ratio (b/a) derived using Galfit, single component Sersic fits in the WFPC2/F300W-band(450 hashing) and the ACS/F850LP band (1350 hashing) for our 162 galaxies. The mean values of the axis ratio are (b/a)F300W = 0.42±0.23and (b/a)F850LP = 0.49 ± 0.22. A two sample KS test shows that there is > 99% probability of the two samples being drawn from adifferent parent population. (right) A scatter plot of the axis ratio b/a(F300W ) vs b/a(F850LP ). Note the paucity of sources withb/a(F850LP ) ≤ 0.2. The average errorbars are derived using the differences in the computed axis-ratio b/a using two different PSFs in theF300W band as shown in Fig. 3.

that in the rest-frame UV, the number of high ellip-ticity (e ≥ 0.8) objects is higher by a factor ∼ 2.8compared to the rest-frame optical. This might explainthe reported claim in literature that high-z LBGs tendto show a significant skew towards higher ellipticities(e.g. Ravindranath et al. 2006). Also, the mean val-ues of the axis ratio are (b/a)F300W = 0.42 ± 0.23 and(b/a)F850LP = 0.49± 0.22. Again, a two sample KS testshows that there is > 99% probability of the two samplesbeing drawn from a different parent population. Fig 7(right) shows a scatter plot of the axis ratio (b/a)F300W

vs (b/a)F850LP . Here we see that in the F300W band,the axis-ratio (b/a) is lower (i.e ellipticity e = (1 − b/a)is higher) in a large number of cases compared to theF850LP band. The average errorbars are derived usingthe differences in the computed axis-ratio b/a using twodifferent PSFs in the F300W band as shown in Fig. 3.Another way to visualize the difference in the derived

quantities n and b/a in UV vs optical is presented inFig. 8. Fig. 8 (left) shows a histogram of the differencebetween the derived values of Sersic index n(F300W )and n(F850LP ) for our sample objects. The smoothcurve is the normal distribution, centered on the ori-gin and having the same variance as the distributionof the sample galaxies. The peak close to zero corre-sponds to objects having same value of n in both thefilters. However the dispersion about the peak is notsymmetrical. Notice the sharp departure from pure gaus-sianity around −1.0 ≤ n(F300W ) − n(F850LP ) ≤ 0.0.These objects correspond to the clump of objects aroundn(F300W ) ≤ 0.5 and n(F850LP ) ∼ 1.0 shown inFig. 6(right). We quantified this by calculating a mea-sure of skewness4 (Fisher Gamma 1) for this distribu-

4 We would like to inform the reader that while we had to binthe data for the purpose of plotting in Fig. 8, the calculation of the

tion, defined as γ1 = m3/m3/22 , where m2 and m3 are

the second and third moments of the distribution respec-tively. We find the value of the skew to be γ1 = −0.10.This negative value of skew γ1 indicates that the dis-tribution has more area under the left tail than ex-pected from a normal distribution. This implies thatn(F300W ) − n(F850LP ) ≤ 0.0 for a larger number ofcases than expected by pure chance (normal distribu-tion), i.e. the Sersic index n is lower in the UV comparedto the optical in a large number of cases. This differ-ence is mainly dominated by galaxies with low values ofn(F300W ) ≤ 0.5 that have a value of n(F850LP ) ∼ 1.0,and is not affected by galaxies with high values of n.Similarly, Fig. 8 (right) shows a histogram of the dif-

ference between b/a(F300W ) and b/a(F850LP ). Again,the smooth curve is the normal distribution, centered onthe origin and having the same variance as the distribu-tion of the sample galaxies. The value of skew in this caseis found to be γ1 = −0.30 (left skewed). Again, a neg-ative value of γ1 implies that the distribution has morearea under the left tail than expected from a normal dis-tribution and that the axis ratio b/a is lower in the UVcompared to the optical in a larger number of cases thanexpected by random chance (normal distribution).Since we are using objects which have a wide range

of photometric redshifts, we are probing different rest-frame wavelengths for different objects with the samefilter. For example, the F300W band will probe rest-frame 2000 A (NUV) for a redshift 0.5 object, whereasthe same filter will probe rest frame 1200 A (FUV) fora redshift 1.5 object. We checked whether this couldbe affecting our interpretation of Fig. 6 & 7 by restrict-ing our sample to only objects with 0.5 ≤ zphoto ≤ 1.0.This yields a sample of 58 objects. Using a calcula-

skew was done from first principles and no binning was used.

7

Fig. 8.— (left) A histogram of the difference between n(F300W ) and n(F850LP ) for our sample galaxies. The smooth curve is thenormal distribution, centered on the origin and having the same variance as the distribution of the sample galaxies. Notice the sharpdeparture from pure gaussianity around −1.0 ≤ n(F300W )− n(F850LP ) ≤ 0.0. These objects correspond to the clump of objects aroundn(F300W ) ≤ 0.5 and n(F850LP ) ∼ 1.0 shown in Fig. 6(b). (right) A histogram of the difference between b/a(F300W ) and b/a(F850LP )for the same sample. Again, the smooth curve is the normal distribution, centered on the origin and having the same variance as thedistribution of the sample galaxies. Both the distributions are found to have negative skew (left skewed), with the distribution having morearea under the left tail than expected from a normal distribution.

tion similar to the one explained above, we find thatnF300W ≤ 0.5 for 22 objects, whereas nF850LP ≤ 0.5 foronly 9 objects. This yields a factor ∼2.4 overestimationof the number of merger candidates in the rest-frameUV (1500 A- 2000 A) compared to the rest-frame optical(4250 A- 5700 A). This compares favorably with the fac-tor ∼2 we had determined using the entire sample of 162objects. This factor is found to be ∼ 1.9, if we employ an ≤ 0.8 as the cutoff for identifying merger candidates,again comparable to the factor we obtained using theentire sample. It is encouraging to see that our resultsare robust and change little irrespective of the fact asto whether or not we apply the redshift cutoff. Part ofthe reason is that the redshift distribution of our sampleobjects anyway peaks between 0.5 and 1.0 (Fig. 1) witha median of 1.02, so that in most cases we are probingrest-frame 1500 A- 2000 A. Note the fact that our resultis stronger (albeit with a smaller sample size) if we useonly objects with 0.5 ≤ zphoto ≤ 1.0. This means thatthe use of objects without regard to their redshift dilutesthe result (as expected) and must therefore be treated asa lower limit.

5. COMMENTS ON INDIVIDUAL SOURCES

We shortlist objects with n(F300W ) ≤ 0.5 and 0.7 ≤

n(F850LP ) ≤ 1.3 to check our assertion that these ob-jects are actually patchy starforming disk dominatedgalaxies, rather than major merger candidates. Thereare 17 objects satisfying the above mentioned criteria.Fig. 9 shows the u− z color map stamps (right) of theseobjects, along with F300W (left) and F850LP (center)imaging graylevel stamps. The color bar in each colormap stamp shows the u − z color range from 0 to 4. Adescription of each object is presented below.

J033233.90-274237.9 This is a smooth face on disk

galaxy as evidenced visually from the F850LP bandimage with a n(F850LP ) = 1.03 and a red nu-cleus seen clearly in the optical image as well asin the u − z color map. There is a blue HII re-gion seen in the F300W band image westwardsfrom the nucleus. It is this patchy star forma-tion which is responsible for a significantly lowern(F300W ) = 0.30. The F850LP band image con-firms that there is no sign of any merger activity.

J033244.13-274234.1 This seems like an edge on diskgalaxy with n(F850LP ) = 0.97. The light distri-bution in the F300W band is more diffuse, leadingto a lower value of n(F300W ) = 0.38.

J033235.98-274217.7 This is a marginal case, withn(F300W ) = 0.47 only slightly smaller thann(F850LP ). The light distribution is slightlymore diffuse in the F300W band compared to theF850LP band. This shows up as blue structureat the North-East edge of the galaxy as seen inthe color map. This might be contributing to themarginally lower value of n(F300W ) compared tothe optical.

J033237.68-274219.3 This is a rather diffuse diskgalaxy, with a distinct nucleus seen in the F850LPband and n(F850LP ) = 0.90. The F300W bandlight distribution is rather patchy with no clear cen-ter evident. This is responsible for the low value ofn(F300W ) = 0.18.

J033236.34-274202.6 This is a patchy star forminggalaxy with several blue HII regions seen clearlyin the F300W band image as well as in thecolor map (particularly one due north-west of the

8

center). The smooth underlying stellar popula-tion is revealed in the F850LP band image withn(F850LP ) = 1.29. The patchy star formation isresponsible for the very low value of n(F300W ).

J033242.23-274130.4 This is a linear chain typegalaxy with n(F850LP ) = 0.71. The light distri-bution is more diffuse in F300W band leading to alower value of n(F300W ). The galaxy as a wholeis rather blue with u− z = 0.38.

J033240.57-274131.2 This is a featureless disk galaxywith n(F850LP ) = 0.91. A nucleus is clearly dis-cernible in the F850LP band, though one is notclearly identifiable in the F300W band. This diffusedistribution of light in the F300W band might beresponsible for the low value of n(F300W ) = 0.25.There is no sign of any merger activity.

J033242.24-274131.1 This is a linear chain typegalaxy. The F300W band light distribution is morediffuse compared to the F850LP band, leading toa lower value of n(F300W ).

J033239.04-274132.4 This objects shows clear evi-dence of patchy star formation in the F300W bandas well as in the color map. Multiple HII regionscan be seen leading to a low value of n(F300W ) =0.16. The F850LP band light distribution is muchsmoother with a close to exponential light profile(n(F850LP ) = 1.03).

J033242.77-274144.6 This is an edge on disk galaxywith n(F850LP ) = 0.81. This is a marginal case,since the n(F300W ) is only slightly smaller thann(F850LP ), owing probably to the slightly morediffuse light distribution in the F300W band.

J033231.49-274158.0 This is a rather compact galaxywith a Sersic profile slightly steeper than an expo-nential n(F850LP ) = 1.28. The F300W band lightdistribution is more diffuse, showing up as a bluehalo conspicuous to the north-west of the galaxycenter. In particular, there is no evidence of anymerger activity.

J033238.34-274120.4 This is an elongated galaxywith n(F850LP ) = 1.05. The patchy light dis-tribution in the F300W band is reminiscent of anirregular, with an ill defined center and parameter-ized by a very low value of n(F300W ).

J033239.25-274059.3 This looks like a diffuse, lowsurface brightness disk galaxy with n(F850LP ) =0.96. A bright HII region (offset from the center)is clearly seen in the F300W band image as well asin the color map, which might be responsible forthe low value of n(F300W ). No sign of any mergeractivity in the F850LP band.

J033241.88-274059.9 This is a rather red (u − z =3.34), compact and featureless galaxy. The F300Wband light distribution is rather diffuse, probablyleading to a low value of n(F300W ). In particular,there are no signs of any merger activity.

J033241.64-274103.4 This is an edge on disk galaxywith n(F850LP ) = 1.19. The F300W band lightdistribution is more fragmented: notice the rela-tively redder center with bluer edges due north-west and south-east of the center in the color map.Again, there are no signs of any merger activity.

J033238.75-274027.8 This is a compact galaxywith a steeper than exponential light profilen(F850LP ) = 1.10. The F300W band light distri-bution is fragmented, leading to a very low valueof n(F300W ).

J033241.42-274044.8 This is a very diffuse galaxy,reminiscent of an Irregular. This is a marginal casewith the n(F300W ) only slightly lower than then(F850LP ).

As can be seen from the individual description aboveand from Fig. 9, most of the objects with n(F300W ) ≤0.5 and n(F850LP ) ∼ 1.0 are probably clumpy starform-ing galaxies with an underlying disk (n ∼ 1.0) of olderstellar population. Many objects are isolated and com-pact and the low value of n(F300W ) can be attributedto diffuse/patchy light distribution in the F300W bandrather than any merger activity. As mentioned earlier,while it is true that blue neighbours will tend to mimicHII regions in the rest-frame UV by producing a shallowSersic profile, however the very fact that they contributelittle or no flux in the optical (parametrized by a steepern(F850LP ) ∼ 1.0) means that they cannot be massivesystems (as z band flux is a better tracer of stellar massthan UV flux) and will constitute minor mergers at best.This analysis should go a long way in convincing thereader that using Sersic profile fitting in the rest-frameUV is fraught with severe biases and should be used withcaution.

6. SUMMARY AND CONCLUSION

We have compared the rest-frame UV vs rest-frameoptical morphologies of intermediate redshift (zmedian =1.02) UV bright galaxies, selected using the deepHST/WFPC2 U-band HUDF parallels image in com-bination with the HST/ACS F850LP GOODS dataset.The UV bright galaxies in this case are at lower redshiftsand have lower luminosities than high-z LBGs (de Melloet al. 2006). We performed single component Sersic fitsin both WFPC2/F300W and ACS/F850LP bands for the162 objects and find that:1) In the rest-frame UV, the Sersic index n is lower in

a large number of cases compared to the rest-frame opti-cal. This is possibly due to the fact that rest-frame UVlight is sensitive to star-forming HII regions which arefragmented and patchy in nature leading to a light dis-tribution which is shallower than the underlying galaxylight distribution seen in the rest-frame optical. This dif-ference is mainly caused by galaxies with low values ofn(F300W ) ≤ 0.5 that have a value of n(F850LP ) ∼ 1.0,and is not affected by galaxies with high values of n, ascan be clearly seen in Fig. 6(right). This has serious con-sequences as the Sersic index n is widely used for identi-fying the morphological class of a galaxy, with n flatterthan exponential being used for identifying merger can-didates. In particular, we find that the use of rest-frameUV overestimates the number of merger candidates by

9

∼40%-100% compared to the rest-frame optical, depend-ing upon the cutoff in n employed for identifying mergercandidates. This shows that the so-called morphologicalK-correction is a serious consideration in constraininggalaxy morphology at intermediate redshifts, as shownby Conselice et al. (2008). Using n ≤ 0.8 as the criterionfor identifying merger candidates, we recommend a cor-rection factor of ∼ 1.4 to the result of Ravindranath etal. (2006) for calculating the fraction of LBGs at z ∼ 3which are likely to be merger candidates.2) We also find that in the rest-frame UV, the number

of high ellipticity (e ≥ 0.8) objects is higher by a factor∼ 2.8 compared to the rest-frame optical. This might ex-plain the reported results in literature that high-z LBGstend to show a skew towards higher ellipticities (Ravin-dranath et al. 2006), as most such work has been doneusing rest-frame UV datasets. We also note that the axisratio b/a is lower, i.e. ellipticity is higher in rest-frameUV compared to the rest-frame optical. The mean val-ues of the axis ratio are (b/a)F300W = 0.42 ± 0.23 and(b/a)F850LP = 0.49± 0.22.This indicates that the reported dominance of elon-

gated morphologies among high-z LBGs might just be abias related to the use of rest-frame UV datasets. How-ever, we cannot conclusively exclude the possibility thatLBGs might still have intrinsically different or elongatedmorphologies.Our results are in agreement with the work of Papovich

et al. (2005), who found that z ∼ 1 galaxies show strong

transformation between their rest-frame UV and opticalmorphologies, using NICMOS observations of HDF-N. Ithas been reported (Papovich et al. 2005, Conselice etal. 2005) that at redshifts 2.0 ≤ z ≤ 3.0, the morphol-ogy of a galaxy remains essentially unchanged from rest-frame UV to rest-frame optical. However, one shouldnot neglect the effect produced by the (1 + z)−4 sur-face brightness dimming, which at higher redshifts sup-presses the (relatively low surface brightness) underly-ing galaxian light (by ∼ 6 magnitudes at z ∼ 3.0), sothat the only parts of a galaxy that are left visible evenin the rest-frame optical are the high surface brightnessHII regions (while the underlying galaxy remains uncon-strained). Future medium depth survey like GOODS(depth ∼10 orbits) using the Near-IR channel of the up-coming camera, WFC3 on HST might fail to probe theunderlying rest-frame optical galaxian light at z ∼ 3.0simply due to (1 + z)−4 surface brightness dimming ef-fects. Ultra-deep images (depth ∼150 orbits) with theNear-IR channel of WFC3 will be required to constrainthe underlying galaxian light at z ∼ 3.0 and will allowsimilar analysis of the morphology of galaxies to high-z.

A.R. would like to thank CSIR for PhD. funding. Spe-cial thanks to Swara Ravindranath for fruitful discussionsand comments. We would like to thank the referee forhis/her insightful comments and constructive criticismsthat have greatly improved the quality of this work.

REFERENCES

Bertin, E., & Arnouts S. 1996, A&AS, 117, 393Conselice, C. J. 2003, ApJS, 147, 1Conselice, C. J., Blackburne, J. A., & Papovich, C. 2005, ApJ, 620,

564Conselice, C. J., Rajgor, S., & Myers, R. 2008, MNRAS, 386, 909de Mello, D. F. et al. 2006, AJ, 131, 216Giavalisco, M. 2002, ARA&A, 40, 579Giavalisco, M. et al. 2004, ApJ, 600, L93Lotz, Jennifer M., Primack, J. & Madau, P. 2004, AJ, 128, 163Papovich, C. et al. 2005, ApJ, 631, 101Peng, C. Y., Ho, L. C., Impey, C. D., & Rix, H. W. 2002, AJ, 124,

266

Ravindranath, S. et al. 2006, ApJ, 652, 963Rawat, A. et al. 2007, A&A, 469, 483Rawat, A. et al. 2008, ApJ, 681, 1089Sersic, J. L., 1968, Atlas de Galaxias Australes(Cordoba: Obs.

Astron, Univ. Nac. Cordoba)Steidel, C.C., et al. 2003, ApJ, 592,728Teplitz, H. I. et al. 2006, AJ, 132, 853Wadadekar, Y., Casertano, S. & de Mello, D.F., 2006, AJ, 132,

1023Wadadekar, Y. et al. 2006, PASP, 118, 450

10

-3 -2 -1 0 1 2 3

-3

-2

-1

0

1

2

3

J033233.90-274237.9 0.57

0.30 22.96

-3 -2 -1 0 1 2 3

-3

-2

-1

0

1

2

3

J033233.90-274237.9 0.57

1.03 21.09

-3 -2 -1 0 1 2 3

-3

-2

-1

0

1

2

3

0

1

2

3

4

-3 -2 -1 0 1 2 3

-3

-2

-1

0

1

2

3

J033244.13-274234.1 0.97

0.38 26.48

-3 -2 -1 0 1 2 3

-3

-2

-1

0

1

2

3

J033244.13-274234.1 0.97

0.97 25.47

-3 -2 -1 0 1 2 3

-3

-2

-1

0

1

2

3

0

1

2

3

4

-3 -2 -1 0 1 2 3

-3

-2

-1

0

1

2

3

J033235.98-274217.7 0.91

0.47 25.47

-3 -2 -1 0 1 2 3

-3

-2

-1

0

1

2

3

J033235.98-274217.7 0.91

0.71 24.22

-3 -2 -1 0 1 2 3

-3

-2

-1

0

1

2

3

0

1

2

3

4

-3 -2 -1 0 1 2 3

-3

-2

-1

0

1

2

3

J033237.68-274219.3 1.52

0.18 26.07

-3 -2 -1 0 1 2 3

-3

-2

-1

0

1

2

3

J033237.68-274219.3 1.52

0.90 23.09

-3 -2 -1 0 1 2 3

-3

-2

-1

0

1

2

3

0

1

2

3

4

Fig. 9.— F300W, F850LP and the u− z color map images of objects with n(F300W ) ≤ 0.5 and 0.7 ≤ n(F850LP ) ≤ 1.3. Explanationis given at the end of this figure.

11

-3 -2 -1 0 1 2 3

-3

-2

-1

0

1

2

3

J033236.34-274202.6 0.04

0.03 24.48

-3 -2 -1 0 1 2 3

-3

-2

-1

0

1

2

3

J033236.34-274202.6 0.04

1.29 21.88

-3 -2 -1 0 1 2 3

-3

-2

-1

0

1

2

3

0

1

2

3

4

-3 -2 -1 0 1 2 3

-3

-2

-1

0

1

2

3

J033242.23-274130.4 1.45

0.07 25.66

-3 -2 -1 0 1 2 3

-3

-2

-1

0

1

2

3

J033242.23-274130.4 1.45

0.71 25.28

-3 -2 -1 0 1 2 3

-3

-2

-1

0

1

2

3

0

1

2

3

4

-3 -2 -1 0 1 2 3

-3

-2

-1

0

1

2

3

J033240.57-274131.2 0.23

0.25 25.51

-3 -2 -1 0 1 2 3

-3

-2

-1

0

1

2

3

J033240.57-274131.2 0.23

0.91 23.32

-3 -2 -1 0 1 2 3

-3

-2

-1

0

1

2

3

0

1

2

3

4

-3 -2 -1 0 1 2 3

-3

-2

-1

0

1

2

3

J033242.24-274131.1 1.73

0.05 25.84

-3 -2 -1 0 1 2 3

-3

-2

-1

0

1

2

3

J033242.24-274131.1 1.73

0.77 24.26

-3 -2 -1 0 1 2 3

-3

-2

-1

0

1

2

3

0

1

2

3

4

Fig. 9.— Continued.

12

-3 -2 -1 0 1 2 3

-3

-2

-1

0

1

2

3

J033239.04-274132.4 0.69

0.16 23.64

-3 -2 -1 0 1 2 3

-3

-2

-1

0

1

2

3

J033239.04-274132.4 0.69

1.03 22.02

-3 -2 -1 0 1 2 3

-3

-2

-1

0

1

2

3

0

1

2

3

4

-3 -2 -1 0 1 2 3

-3

-2

-1

0

1

2

3

J033242.77-274144.6 0.72

0.41 25.55

-3 -2 -1 0 1 2 3

-3

-2

-1

0

1

2

3

J033242.77-274144.6 0.72

0.81 23.51

-3 -2 -1 0 1 2 3

-3

-2

-1

0

1

2

3

0

1

2

3

4

-3 -2 -1 0 1 2 3

-3

-2

-1

0

1

2

3

J033231.49-274158.0 0.52

0.41 25.61

-3 -2 -1 0 1 2 3

-3

-2

-1

0

1

2

3

J033231.49-274158.0 0.52

1.28 23.68

-3 -2 -1 0 1 2 3

-3

-2

-1

0

1

2

3

0

1

2

3

4

-3 -2 -1 0 1 2 3

-3

-2

-1

0

1

2

3

J033238.34-274120.4 1.80

0.06 26.32

-3 -2 -1 0 1 2 3

-3

-2

-1

0

1

2

3

J033238.34-274120.4 1.80

1.05 24.88

-3 -2 -1 0 1 2 3

-3

-2

-1

0

1

2

3

0

1

2

3

4

Fig. 9.— Continued.

13

-3 -2 -1 0 1 2 3

-3

-2

-1

0

1

2

3

J033239.25-274059.3 0.79

0.33 24.74

-3 -2 -1 0 1 2 3

-3

-2

-1

0

1

2

3

J033239.25-274059.3 0.79

0.96 22.43

-3 -2 -1 0 1 2 3

-3

-2

-1

0

1

2

3

0

1

2

3

4

-3 -2 -1 0 1 2 3

-3

-2

-1

0

1

2

3

J033241.88-274059.9 0.73

0.02 27.19

-3 -2 -1 0 1 2 3

-3

-2

-1

0

1

2

3

J033241.88-274059.9 0.73

0.84 23.85

-3 -2 -1 0 1 2 3

-3

-2

-1

0

1

2

3

0

1

2

3

4

-3 -2 -1 0 1 2 3

-3

-2

-1

0

1

2

3

J033241.64-274103.4 1.81

0.32 26.40

-3 -2 -1 0 1 2 3

-3

-2

-1

0

1

2

3

J033241.64-274103.4 1.81

1.19 23.34

-3 -2 -1 0 1 2 3

-3

-2

-1

0

1

2

3

0

1

2

3

4

-3 -2 -1 0 1 2 3

-3

-2

-1

0

1

2

3

J033238.75-274027.8 1.39

0.06 26.73

-3 -2 -1 0 1 2 3

-3

-2

-1

0

1

2

3

J033238.75-274027.8 1.39

1.10 24.39

-3 -2 -1 0 1 2 3

-3

-2

-1

0

1

2

3

0

1

2

3

4

Fig. 9.— Continued.

14

-3 -2 -1 0 1 2 3

-3

-2

-1

0

1

2

3

J033241.42-274044.8 1.48

0.30 25.68

-3 -2 -1 0 1 2 3

-3

-2

-1

0

1

2

3

J033241.42-274044.8 1.48

0.75 24.42

-3 -2 -1 0 1 2 3

-3

-2

-1

0

1

2

3

0

1

2

3

4

Fig. 9.— F300W, F850LP and the u− z color map images of objects with n(F300W ) ≤ 0.5 and 0.7 ≤ n(F850LP ) ≤ 1.3. In each row,the left cutout is a F300W filter HST/WFPC2 image from the Hubble-UDF parallels, the center cutout is the F850LP HST/ACS image ofthe same object from the GOODS survey, while the right cutout is the u− z color map. Labels in the F300W and F850LP band cutoutsare the same as in Fig. 2. Each image is 6 arcsec × 6 arcsec. North is up, East is to the left.